Attribute inductive data analysis

Abstract

A method and apparatus for monitoring a physical process comprising a plurality of interacting attributes where the attributes are components of the physical process. The method and apparatus locates defective attributes and defective interactions between interacting attributes within the physical process. The apparatus comprises a processor, and input member and a hierarchical data structure. Data concerning the attributes of the physical process are input and organized into the hierarchical data structure. A response variable and a variation in the response variable for each population in the hierarchical data structure is determined, and used to identify defects in the physical process.

I claim: 
     
       1. A computer based apparatus for monitoring a physical process and enabling identification of a source of defect in the physical process, the physical process comprising a plurality of interacting attributes, the attributes being components of the physical process, said plurality of interacting attributes and said interactions between said plurality of interacting attributes each having in ideal condition a predetermined desired state, wherein the apparatus locates defective ones of the plurality of interacting attributes and defective interactions between the plurality of interacting attributes within the physical process to enable identification of the source of defect in the physical process for subsequent physical adjustment of the source of defect, the apparatus comprising: a. means for monitoring said plurality of interacting attributes and said interactions between said plurality of interacting attributes and generating a data set comprising instances of data concerning each one of the plurality of interacting attributes and the interactions between the plurality of interacting attributes of the physical process, said instances of data comprising instances of deviation of said plurality of interacting attributes and said interactions between said plurality of interacting attributes from their respective predetermined desired states;   b. a hierarchical data structure comprising populations which correspondingly represent one of the plurality of interacting attributes of the physical process or one of the interactions between the plurality of interacting attributes of the physical process, the hierarchical data structure comprising: (i) a first level including a universe population which is a union of all populations,   (ii) a second level including a first plurality of populations which correspondingly represent one of the plurality of interacting attributes of the physical process, and   (iii) a plurality of subsequent levels including a second plurality of populations representative of the interactions between the plurality of interacting attributes, wherein the second plurality of populations at each one of the plurality of subsequent levels correspondingly represent an additional interaction between the plurality of interacting attributes than at a previous one of the plurality of subsequent levels;     c. means, coupled to the hierarchical data structure and to the monitoring means, for substituting the data set into the hierarchical data structure; and   d. means, coupled to the hierarchical data structure, including (i) means for identifying a response variable for each population, the response variable for a population being a representation of the operation of the interacting attributes of the population in relation to an ideal operation of the interacting attributes in a physical process where no defects occur,   (ii) means for comparing the response variable of each population with the response variable of other populations in the hierarchical data structure,   (iii) means for determining a variation in the response variable for each population, and   (iv) means for identifying a defective population to be a population having the most significant variation in its response variable, significance being determined by a first preselected scheme such that the identified population corresponds to the individual one of the plurality of interacting attributes or an interaction between the plurality of interacting attributes which is the most likely source of a defect within the physical process.     
     
     
       2. The apparatus of claim 1 further comprising: a. means for storing an identifying reference as a first identified population to the population in the hierarchical data structure corresponding to the individual one of the plurality of interacting attributes or the interaction between the plurality of interacting attributes that is the most likely source of a defect within the physical process;   b. means for determining whether the first identified population is a significant population, significance being determined according to a second preselected scheme;   c. means for creating a revised data set when the first identified population is not a significant population by revising the data in the data set to eliminate the variation in the response variable of the first identified population; and   d. means for identifying a second identified population in the revised data set, the second identified population corresponding to an individual one of the plurality of interacting attributes or an interaction between the interacting attributes in the physical process that is defective independent of the first identified population and has the most significant variation in its response variable, significance determined by the first preselected scheme.   
     
     
       3. The apparatus of claim 1 further comprising: a. means for selecting as a test population the universe population;   b. means for identifying an identified population at a subsequent level of the hierarchical data structure to the test population that has the most significant variation in the response variable, significance being determined by the first preselected scheme;   c. means for determining whether the identified population is a significant population, significance being determined according to a second preselected scheme;   d. means for determining whether the test population is equal to the universe population;   e. means for selecting as a new test population the identified population if the identified population is a significant population;   f. means for determining when all of the populations in the hierarchical data structure at levels subsequent to the test population do not have a significant variation in their response variable, significance being determined by the first preselected scheme;   g. means for halting the computer based apparatus when the test population is equal to the universe population; and   h. means for outputting the identified population having a significant variation in its response variable and identified as a significant population as the population representative of the individual one of the plurality of interacting attributes or the interaction between the plurality of interacting attributes which is the most likely source of a defect within the physical process.   
     
     
       4. The apparatus of claim 2 or 3 wherein the second preselected scheme comprises determining whether the variation in the response variable of a population is greater than a threshold value, the threshold value being a predetermined value representative of a level of risk that the population may be more defective than other populations. 
     
     
       5. The apparatus of claim 1 wherein the data set is derived from a manufacturing process. 
     
     
       6. The apparatus of claim 1 wherein each one of the plurality of interacting attributes are parameters from a manufacturing process. 
     
     
       7. The apparatus of claim 1 wherein the response variable for each population is a fraction defective, the fraction defective being the total number of defects in the population divided by the total number of opportunities for defects in the population. 
     
     
       8. The apparatus of claim 1 wherein the variation in the response variable is a known statistical approximation measuring the fraction defective of a population relative to the fraction defective of the population's complement, the complement of a population being that portion of the data set that remains when the population is removed from the data set. 
     
     
       9. The apparatus of claim 1 wherein the first preselected scheme comprises selecting the population with the largest variation in its response variable. 
     
     
       10. A device for monitoring a physical process and locating defective ones of a plurality of interacting attributes and interactions between the plurality of interacting attributes of the physical process to enable identification of the source of defect in the physical process for subsequent physical adjustment of the source of defect, said plurality of interacting attributes and said interactions between said plurality of interacting attributes each having in ideal condition a predetermined desired state, the device comprising: a1. means for monitoring said plurality of interacting attributes and said interactions between said plurality of interacting attributes and generating an attribute characterized data set comprising instances of data concerning each one of said plurality of interacting attributes and said interactions between said plurality of interacting attributes of the physical process, said instances of data comprising instances of deviation of said plurality of interacting attributes and said interactions between said plurality of interacting attributes from their respective predetermined desired states;   a2. means for receiving as input said attribute characterized data set comprising instances of data concerning each one of the plurality of interacting attributes and the interactions between the plurality of interacting attributes of the physical process;   b. a hierarchical data structure comprising populations which correspondingly represent one of the plurality of interacting attributes of the physical process or one of the interactions between the plurality of interacting attributes of the physical process, the hierarchical data structure comprising: (i) a first level including a universe population which is a union of all populations,   (ii) a second level including a first plurality of populations which correspondingly represent one of the plurality of interacting attributes of the physical process, and   (iii) a plurality of subsequent levels including a second plurality of populations representative of the interactions between the plurality of interacting attributes, wherein the second plurality of populations at each one of the plurality of subsequent levels correspondingly represent an additional interaction between the plurality of interacting attributes than at a previous one of the plurality of subsequent levels;     c. means coupled to the means for receiving for organizing the attribute characterized data set into a hierarchical data structure;   d. means for identifying a response variable for each population, the response variable for a population being a numerical representation of the operation of the attributes comprising the population in relation to an ideal operation of the attributes in a physical process where no defects occur; and   e. a processor, coupled to the hierarchical data structure, for selecting defect populations in the hierarchical data structure, wherein a defect population is the population that has a response variable that identifies it as representing one of the plurality of interacting attributes or combination of interacting attributes that is a likely source of a defect within the physical process.   
     
     
       11. The device of claim 10 wherein the means for selecting defect populations further comprises: a. means for selecting as a test population the universe population;   b. means for determining if the test population is equal to the universe population;   c. means for determining a variation in the response variable for each population;   d. means for identifying an identified population at a subsequent level of the hierarchical data structure to the test population that has the most significant variation in its response variable, significance being determined by a scheme that selects the population with the largest variation in its response variable;   e. means for selecting as a new test population the identified population if the variation in the response variable of the identified population is greater than a threshold value, the threshold value being a predetermined value representative of the level of risk that the identified population is more defective than other populations in the hierarchical data structure;   f. means for storing in a list of isolated sources of variations a reference to the identified population if the variation in its response variable is not greater than the threshold value;   g. means for creating a revised data set when the variation in the response variable of the identified population is not greater than the threshold value by revising the data in the data set to eliminate the variation in the response variable associated with the identified population;   h. means for reselecting as the test population the universe population when the variation in the response variable of the identified population is not greater than the threshold value;   i. means for determining when all of the populations in the hierarchical data structure at subsequent levels to the test population do not have a significant variation in their response variable;   j. means for terminating the operation of the device when the test population is equal to the universe population; and   k. means for outputting the populations stored in the list of isolated sources of variations, the populations representative of the interacting attributes and interactions between the plurality of interacting attributes which are the most likely sources of defect within the physical process.   
     
     
       12. The device of claim 10 wherein the interacting attributes are interacting components in a manufacturing process. 
     
     
       13. A method for monitoring a physical process and identifying defective combinations of components in the physical process to enable subsequent physical adjustment of the source of defect in the physical process, the physical process comprising a plurality of interacting components, the components being represented by attributes of the physical process, said attributes each having in ideal condition a predetermined desired state, the method comprising the steps of: a. monitoring said physical process and gathering a data set comprising instances of data concerning each one of a plurality of interacting attributes of the physical process and interactions between the plurality of interacting attributes of the physical process, said instances of data comprising instances of deviation of said plurality of interacting attributes and said interactions between said plurality of interacting attributes from their respective predetermined desired states;   b. organizing the data set into a hierarchical data structure comprising populations which correspondingly represent one of the plurality of interacting attributes of the physical process or one of the interactions between the plurality of interacting attributes of the physical process, the hierarchical data structure comprising: (i) a first level including a universe population which is a union of all populations,   (ii) a second level including a first plurality of populations which correspondingly represent one of the plurality of interacting attributes of the physical process, and   (iii) a plurality of subsequent levels including a second plurality of populations representative of the interactions between the plurality of interacting attributes, wherein the second plurality of populations at each one of the plurality of subsequent levels correspondingly represent an additional interaction between the plurality of interacting attributes than at a previous one of the plurality of subsequent levels;     c. identifying a response variable for each population, the response variable for a population being a numerical representation of the operation of the interacting components of the population in relation to an operation of the interacting components in a physical process where no defects occur;   d. comparing the response variable of each population with the response variable of the other populations in the hierarchical data structure;   e. determining a variation in the response variable for each population; and   f. identifying a defective population in the hierarchical data structure as that population representative of one of the plurality of interacting attributes or one of the interactions between the interacting attributes that is the most likely source of defect within the physical process.   
     
     
       14. The method of claim 13 wherein the step of identifying the defective population further comprises the steps of: a. selecting as a test population the universe population;   b. identifying an identified population at a subsequent level of the hierarchical data structure to the test population that has the most significant variation in the response variable, significance being determined by a preselected scheme;   c. selecting as a new test population the identified population if the variation in the response variable of the identified population is greater than a predetermined threshold value which is a predetermined value representative of the level of risk that the identified population may be more defective than other populations at the same level in the hierarchical data structure;   d. revising the data set if the variation in the response variable of the identified population is not greater than the threshold value;   e. determining when all of the populations at a level subsequent to the test population do not have a significant variation in their response variable;   f. determining if the test population is equal to the universe population; and   g. terminating processing when the test population is the universe population.   
     
     
       15. The method of claim 14 wherein the step of revising the data set further comprises the steps of: a. storing an identifying reference to the identified population in the hierarchical data structure;   b. revising the data set to eliminate the effects of the variation in the response variable associated with the identified population;   c. revising the hierarchical data structure to represent the revised data set;   d. selecting as the test population the universe population; and   e. identifying a second defective population in the revised hierarchical data structure being that population which, when examining the variation in the response variable for each population, is the most likely source of a defect within the physical process.   
     
     
       16. The method of claim 14 wherein the step of identifying the identified population further comprises the steps of: (i) determining a variation variable for each population which is a known statistical approximation measuring the response variable of a population relative to the response variable of the population's complement, the complement of a population being that portion of the data set that remains when the population is removed from the data set,   (ii) determining a variation population at a subsequent level of the hierarchical data structure to the test population that has the largest variation variable, and   (iii) identifying the variation population as the identified population.   
     
     
       17. The method of claim 13 wherein the data set is derived from a manufacturing process. 
     
     
       18. The method of claim 13 wherein the attributes are parameters of a manufacturing process. 
     
     
       19. The method of claim 13 wherein the response variable for each population is a fraction, the fraction being the total number of defects in the population divided by the total number of opportunities for defects in the population. 
     
     
       20. The method of claim 13 wherein the step of determining the variation in the response variable for each population comprises the step of determining a variation variable for each population which is a known statistical approximation measuring the response variable of a population relative to the response variable of the population's complement, the complement of a population being that portion of the data set that remains when the population is removed from the data set. 
     
     
       21. A method for monitoring a physical process which comprises a plurality of interacting attributes to determine whether any of said interacting attributes or interactions between said interacting attributes are behaving inefficiently so as to require alteration of the physical process, said plurality of interacting attributes and said interactions between said plurality of interacting attributes each having in ideal condition a predetermined desired state, the method comprising the steps of: a1. monitoring said physical process and generating a data set comprising data representative of the behavior of said plurality of interacting attributes and the behavior of said interactions between said plurality of interacting attributes, said representative data comprising instances of deviation of said plurality of interacting attributes and said interactions between said plurality of interacting attributes from their respective predetermined desired states;   a2. storing said data set from step a1:   a3. retrieving as input said data set comprising data representative of the behavior of the plurality of interacting attributes and the behavior of the interactions between the plurality of interacting attributes;   b. organizing the data set into a hierarchical data structure comprising populations which correspondingly represent one of the plurality of interacting attributes of the physical process or one of the interactions between the plurality of interacting attributes of the physical process, the hierarchical data structure comprising: (i) a first level including a universe population which is a union of all populations,   (ii) a second level including a first plurality of populations which correspondingly represent one of the plurality of interacting attributes of the physical process, and   (iii) a plurality of subsequent levels including a second plurality of populations representative of the interactions between the plurality of interacting attributes, wherein the second plurality of populations at each one of the plurality of subsequent levels correspondingly represents an additional interaction between the plurality of interacting attributes than at a previous one of the plurality of subsequent levels;     c. selecting as a test population the universe population;   d. calculating a statistical approximation value for each population at levels subsequent to the test population, the statistical approximation value being a known statistical approximation measuring the probability that a population has a greater variation than the complement of the population, a population's complement being that portion of the data set that remains when the population is removed from the data set;   e. identifying an identified population at a subsequent level to the test population that has the largest statistical approximation value;   f. determining whether the identified population is a significant population, significance being determined according to a pre-selected scheme;   g. if the identified population is a significant population, (i) selecting as the test population the identified population, and   (ii) repeating steps (c) through (f) until it is determined that the identified population is not a significant population;     h. if the identified population is not a significant population and the test population is not the universe population, (i) altering the data in the data set so that one of the plurality of interacting attributes or interactions between interacting attributes represented by the identified population does not have the largest statistical approximation value,   (ii) adding the test population to a list of isolated sources of variations, and   (iii) repeating steps (c) through (g) until it is determined that the test population is the universe population;     i. outputting the identities of the test populations in the list of isolated sources of variations; and   j. using the test populations in the list of isolated sources of variations to determine whether any attributes or interactions between interacting attributes are behaving inefficiently so as to require alteration of the physical process.   
     
     
       22. The method of claim 21 wherein the data set is derived from a manufacturing process. 
     
     
       23. The method of claim 21 wherein the step of determining whether a population is a significant population further comprises the steps of: (i) taking as input a predetermined threshold value,   (ii) comparing the statistical approximation value of the population with the predetermined threshold value, and   (iii) determining that the population is a significant population if the statistical approximation value of the population is greater than the predetermined threshold value.   
     
     
       24. For use in optimizing the operation of a physical process comprising a plurality of interacting attributes whose behavior affects the operation of the physical process, a computer-based method for monitoring and identifying individual ones of the plurality of interacting attributes and interactions between the plurality of interacting attributes of the physical process that vary from a preselected optimized state and require modification to optimize the operation of the physical process, the method comprising of the steps of: a1. monitoring said physical process and generating a data set comprising data representative of the behavior of said plurality of interacting attributes and the behavior of said interactions between said plurality of interacting attributes;   a2. storing said data set from step a1;   a3. retrieving as input said data set comprising data representative of the behavior of the plurality of interacting attributes and the behavior of the interactions between the plurality of interacting attributes;   b. organizing the data set to a hierarchical data structure comprising populations which correspondingly represent one of the plurality of interacting attributes of the physical process or one of the interactions between the plurality of interacting attributes of the physical process, the hierarchical data structure comprising: (i) a first level including a universe population which is a union of all populations,   (ii) a second level including a first plurality of populations which correspondingly represent one of the plurality of interacting attributes of the physical process, and   (iii) a plurality of subsequent levels including a second plurality of populations representative of the interactions between the plurality of interacting attributes, wherein the second plurality of populations at each one of the plurality of subsequent levels correspondingly represents an additional interaction between the plurality of interacting attributes than at a previous one of the plurality of subsequent levels;     c. selecting as a test population the universe population;   d. calculating a z-complement value for each population at a level subsequent to the test population, the z-complement value being a known statistical approximation measuring the probability that a population has a greater variation in behavior than the complement of the population, a population's complement being that portion of the data set that remains when the population is removed from the data set;   e. identifying an identified population at a level subsequent to the test population that has the largest z-complement value;   f. determining whether the identified population is a significant population, significance being determined according to a pre-selected scheme;   g. if the identified population is a significant population, (i) selecting as the test population the identified population, and   (ii) repeating steps (d) through (g) until it is determined that the identified population is not a significant population;     h. if the identified population is not a significant population and the test population is not the universe population, (i) altering the data in the data set so that one of the plurality of interacting attributes or interactions between interacting attributes represented by the identified population do not vary from the preselected optimized state,   (ii) adding the test population to a list of isolated sources of variations, and   (iii) repeating steps (c) through (h) until it is determined that the test population is the universe population; and     i. outputting the identities of the significant population which correspond to an identified one of the plurality of interacting attributes or an identified one of the interactions between the interacting attributes.   
     
     
       25. The method of claim 24 wherein the data set is derived from a manufacturing process. 
     
     
       26. The method of claim 24 wherein the step of determining whether a population is a significant population further comprises the steps of: (i) taking as input a predetermined threshold value,   (ii) comparing the z-complement value of the population with the predetermined threshold value, and   (iii) determining that the population is a significant population if the z-complement value of the population is greater than the predetermined threshold value.